Follow the link to download the full summary: bcf.princeton.edu/events/serg...
On Thursday, March 28, Sergio Correia join ed Markus’ Academy for a conversation on, “Unlocking Economic Data with LLM.” Sergio Correia is an economist at the Board of Governors of the Federal Reserve System in the Division of Financial Stability.
● A summary in four bullets
○ Although barriers to entry of using Large Language Models (LLMs) appear high, in reality using these tools only requires a few lines of Python, and one need not be a professional coder to do so.
○ The talk covered simple examples that leverage LLMs to summarize publicly available text, extract historical data, or merge datasets. It also covered the basics of embeddings, and showed how these can be useful even without LLMs to classify text or run regressions (for example to predict Amazon product scores based on user reviews)
○ Lastly, the talk covered the nuances around designing a good Retrieval-Augmented Generation (RAG) pipeline to generate data with LLMs
○ All of the code from the talk’s examples can be reviewed here and here
Timestamps:
[0:00] Introduction and poll questions
[3:55] Why should we use AI to create data?
[25:50] The basics of embeddings
[47:02] LLMs and GPT
[58:15] Designing a good RAG is difficult
15 июн 2024
